Boston Dynamics' four-legged Spot robot can now accurately read analog thermometers and pressure gauges in industrial settings. This advancement is powered by Google DeepMind's new Gemini Robotics-ER 1.6 AI model, designed to improve robotic interaction with physical environments through 'embodied reasoning'.
The Gemini Robotics-ER 1.6 model acts as a high-level reasoning system, enabling robots to plan and execute tasks. It also allows for precise reading of complex instruments and visual inspections via sight glasses. This upgrade stems from Google DeepMind's ongoing partnership with Boston Dynamics.
Boston Dynamics is testing its quadruped and humanoid robots in various industrial facilities, including those owned by its parent company, Hyundai Motor Group. Spot is being trialled as a robotic inspector capable of interpreting multiple needles, liquid levels, and text on instruments, requiring sophisticated visual reasoning.
The Gemini Robotics-ER 1.6 model features 'agentic vision,' combining visual reasoning with code execution for a 'visual scratchpad.' This capability significantly boosts robotic performance on instrument reading tasks, with the new model achieving 98 percent accuracy, a jump from the 23 percent of its predecessor.
Even without agentic vision, the baseline Gemini Robotics-ER 1.6 model achieves 86 percent accuracy in instrument reading by processing visual elements and improving 'multi-view reasoning' with multiple camera streams.